|
Cascading is a particular case of ensemble learning based on the concatenation of several classifiers, using all information collected from the output from a given classifier as additional information for the next classifier in the cascade. Unlike voting or stacking ensembles, which are multiexpert systems, cascading is a multistage one. The first cascading classifier is the face detector of Viola and Jones (2001). The requirement was that the classifier be fast in order to be implemented on low CPU systems, such as cameras and phones. == Basic algorithm == The algorithm can be summed up as follows: for all rectangles in the image: * stage 1: is there a face in the current rectangle according to classifier 1 ? if yes stage2, if no rectangle does not contain a face * stage 2: is there a face in the current rectangle according to classifier 2 ? if yes stage3, if no rectangle does not contain a face ... * stage n: is there a face in the current rectangle according to classifier n ? if yes there is a face in current rectangle The question 'is there a face in the current rectangle according to classifier k?', for the stage number k ranging from 1 to n is answered by a collection of weak learners (that is, simple rules that cannot alone do the classification, but are expressive enough to be able to classify anything when combined). For example, the 'face in current rectangle' score of stage 1 might be 0.3 *(centerpixels-leftpixels-rightpixels)+0.5 *(-pixel 1/3 from the top + rest of pixels) > 0.8. The first condition expresses the fact that the nose catches the light (the center will be light and the sides dark, giving a high numerical result) and the second one the fact that eyebrows are darker than the rest of the face. 抄文引用元・出典: フリー百科事典『 ウィキペディア（Wikipedia）』 ■ウィキペディアで「Cascading classifiers」の詳細全文を読む スポンサード リンク
|